Data-Limited Deep Learning Methods for Mild Cognitive Impairment Classification in Alzheimer's Disease Patients.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Mild Cognitive Impairment (MCI) represents a transitional stage between normal cognitive function and dementia, such as Alzheimer's disease (AD).
• Early diagnosis of MCI is crucial for identifying individuals with early signs of AD, potentially enabling timely interventions.
• Magnetic Resonance Imaging (MRI) is a key tool used by initiatives like the Alzheimer's Disease Neuroimaging Initiative (ADNI) for diagnosing MCI and AD.

Purpose of the Study:

• To perform binary classifications between healthy controls (CN) and two types of MCI (Early MCI - EMCI, and Late MCI - LMCI) using limited MRI images.
• To implement and compare the performance of two distinct Convolutional Neural Network (CNN) architectures for MCI classification.
• To evaluate the effectiveness of deep learning approaches in differentiating MCI subtypes from normal cognition based on MRI data.

Main Methods:

• Utilized MRI scans from 516 patients: 172 CN, 172 EMCI, and 172 LMCI.
• Employed deep learning, specifically two different CNN architectures, for binary classification tasks.
• Split the dataset into 50% for training, 20% for validation, and 30% for testing.

Main Results:

• The best classification performance was achieved between CN and LMCI using the coronal view with an accuracy of 79.67% for one CNN model.
• The second proposed CNN model achieved 67.85% accuracy for the same CN vs. LMCI classification using the coronal view.
• These results indicate varying degrees of success in differentiating MCI subtypes from healthy controls depending on the CNN architecture and MRI view.

Conclusions:

• Deep learning models, particularly CNNs, can be applied to MRI data for the classification of MCI subtypes.
• The coronal view of MRI appears to be a significant factor in achieving higher classification accuracy for distinguishing LMCI from CN.
• Further research and model optimization are warranted to improve the accuracy and reliability of AI-driven MCI diagnosis from neuroimaging.